Training method and device of illegal audio recognition model and illegal audio recognition method

• Main Authors: ZHANG ZHENGTONG, XIE RUI, WU WENLIANG, LIAO YANBING, CHEN XIXING, MA FEI, CHEN GUANGYAO, PAN ZISHENG, HUANG XIANGKANG, XU ZHIJIAN, WANG WEIZHE, DENG QICHUN, ZENG RUIHONG, XIONG JIA, LI ZIJUN, JIAO NANKAI, LAN XIANG, MA JINLONG
• Format: Patent
• Language: Chinese; English
• Published: 08.03.2024
• Subjects: ACOUSTICS; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

The invention provides a training method and device of a violation audio recognition model, a violation audio recognition method, computer equipment and a storage medium. The training method comprises the following steps: acquiring a training data set; training the first initial model by using the training data set to obtain a first recognition model; taking the first recognition model as supervision, and training the second initial model by using the training data set to obtain a second recognition model; the parameters in the second recognition model are less than those in the first recognition model, and the second recognition model is used as a violation audio recognition model to be issued to the target mobile terminal so as to locally perform violation audio recognition at the target mobile terminal. The model trained by using the scheme has similar prediction capability to the model of the server side, but can be deployed on equipment with limited computing power for local identification due to fewer p